pocketvectordb 1.0.1

Fast personal vector database with persistent storage and semantic search

PocketVectorDB

A lightweight, fast, offline-ready vector database for Python and mobile/edge environments

PocketVectorDB is a minimal, dependency-light vector database designed for:

• Offline AI agents
• Termux (Android) environments
• Edge devices & IoT
• Local LLM memory systems
• Developers who want fast semantic search without massive dependencies

Built for simplicity and speed, PocketVectorDB stores embeddings in a compact NumPy matrix, supports metadata filtering, batch inserts, persistent storage, and cosine similarity search — all in under 10KB of Python code.

====================================================================== 🚀 Features

• Ultra-lightweight: no server, no heavy frameworks
• Fast cosine similarity search via optimized matrix operations
• Persistent storage using embeddings.npy and metadata.json
• Batch insert operations
• Metadata filtering (where={...})
• Full CRUD operations
• Zero external dependencies except NumPy
• Works on Termux, Linux, macOS, and Windows
• Perfect for small AI agents and local LLM memory

====================================================================== 📦 Installation

Python

pip install pocketvectordb

Termux

Install from a local wheel:

pip install pocketvectordb-1.0.0-py3-none-any.whl

====================================================================== 🧠 Quick Start

from pocketvectordb import VectorDB
import numpy as np

db = VectorDB("./my_vectordb", dimension=384)

embedding = np.random.randn(384).astype(np.float32)
doc_id = db.add(embedding, "Hello world!", metadata={"tag": "greeting"})

query = np.random.randn(384).astype(np.float32)
results = db.query(query, n_results=3)

print(results["documents"])

====================================================================== 🔍 Similarity Search

results = db.query(query_embedding, n_results=5)

Returns:

• ids
• documents
• distances
• metadatas

====================================================================== 🗂️ Metadata Filtering

results = db.query(
    query_embedding,
    n_results=3,
    where={"category": "science"}
)

====================================================================== ✏️ Updating Documents

db.update(
    doc_id,
    text="Updated content",
    metadata={"updated": True}
)

====================================================================== 🗑️ Delete Matching Documents

deleted_count = db.delete(where={"category": "tech"})

====================================================================== 💾 Persistence

Every write updates:

• embeddings.npy
• metadata.json

Reloading is automatic:

db = VectorDB("./my_vectordb")
print(db.count())

====================================================================== 📊 Performance

Benchmarks on Android (Termux):

Documents	Insert Time	Query Time
100	0.017s	0.14 ms
500	0.39s	0.31 ms
1000	1.56s	0.53 ms

PocketVectorDB is optimized for fast local lookups without GPU or FAISS.

====================================================================== 🛠️ Why PocketVectorDB?

Most vector databases are:

• too heavy
• server-based
• cloud-locked
• not optimized for mobile
• overkill for small agents

PocketVectorDB is a pure-Python, offline-first, minimalist vector engine.

Perfect for:

• LLM memory systems
• Personal agents
• Offline chatbots
• IoT classification
• Fast lookup systems
• Mobile AI experiments
• Students & researchers

====================================================================== ❤️ Author

ThatFkrDurk561

PocketVectorDB is built for developers who want speed, portability, and simplicity without massive frameworks.